Are Teachers the AI Police? How Do They Judge Authentic Student Writing?

By Honglin Chen

The rules for school-based assessment for the Higher School Certificate (HSC) changed dramatically on Monday.

Deputy Premier and the Minister for Education Prue Car, announced the change as a measure for protecting the integrity of learning, stating “(w)e cannot be sure that AI is not being used in take-home assessments”.  It starts next term. Now an  HSC course may involve only one take-home assessment task worth no more than 15% of the school-based component (about 7.5% of a student’s final HSC mark).

The new HSC rule is about more than take-home tasks

Crucially, NESA has advised schools not to rely on AI detection tools as their main safeguard against misuse. If used at all, detection software should be treated as “only one input for teacher professional review, not as sole proof of AI use or malpractice”.

This is a welcome and important move. Research tells us AI detection tools are not reliable and can be biased against writers who use English as an additional language (EAL). They are designed to identify patterns associated with linguistically predictable writing.

In particular, they target features like a limited range of vocabulary, repetitive text and paragraph structures, and language which doesn’t vary much. As a result, EAL students’ writing is more likely to be incorrectly flagged as AI-generated. This is because second language learners tend to stick to well-rehearsed language patterns before they develop confidence to innovate more with their language.

Judging authenticity – a new responsibility for teachers

Rejecting detection tools as the main judge of authenticity does not, however, eliminate the challenge of maintaining assessment integrity. It shifts responsibility back to teachers, who are now being asked to exercise professional judgement about whether a piece of writing represents a student’s own thinking and learning.

This is no small task. A NESA-commissioned survey of nearly 3,400 teachers and over 750 school leaders across NSW found that many secondary teachers already rely on their own judgement to evaluate possible AI use. Teachers, however, describe the process as difficult and time-consuming. Half also reported concerns about their ability to prevent AI-related misuse.

Yet much of the public discussion continues to frame this judgement as a matter of spotting AI-like writing through surface features such as overused vocabulary (for example, “genuinely”), formulaic sentence structures (say “not X but Y”), or unusual length. As with AI detection software, relying on these surface features risks inaccurate and inequitable judgements of authenticity.

The real challenge is evaluating learning in writing

So the real challenge in judging writing authenticity is not simply determining whether a piece of writing sounds like AI. A more important question is whether it provides evidence of learning. A checklist of generic ‘Al tells’ is not sufficient. Teachers need a research-informed understanding of what student writing reveals about their learning and thinking.

Research in disciplinary literacy has long examined how students construct knowledge, develop arguments, and position themselves as readers and writers.

What student writing can tell us about thinking and learning

Our research shows that teachers can be supported to notice specific evidence in student writing. Evidence of deep, authentic learning can be seen when students write in disciplinary ways. For example, how the students

• engage with overarching concepts and ideas from their subjects. Writing reflects subject-specific ways of thinking about the world and each subject has different core concerns. This means that students need to write in particular ways, about particular things and using particular language in each of their subjects. Authenticity can be judged through how students’ writing reflects the key ideas, questions and values central to each subject.
• organise information into a coherent whole that meets the demands of the writing task. This requires attending to language choices at the levels of whole text, paragraph and sentence according to the genre, topic and audience. All of this can be taught explicitly and practised in the context of writing tasks that have fidelity to the core concerns of each subject.
• express stance and authorial voice. In many subjects, students must be able to take a clear position, in which they use and evaluate evidence. The types of evidence they draw on and the way they use this evidence to support an argument or explanation is an indicator of their disciplinary expertise.  Authenticity can be judged when students adopt a disciplinary appropriate authorial voice for the task and subject-matter at hand.
• develop across multiple drafts of a task.  Students begin with an initial draft when a topic is introduced. As a unit unfolds, students revisit this draft adding detail and refining the text as knowledge about content and language develops. Authenticity is easier to judge when writing is taken as part of an ongoing learning journey.

An opportunity to rethink how we teach and assess writing

Generative AI has unsettled familiar assumptions about student writing. At the same time, it has also sharpened what is at stake. Writing is not simply the production of polished written products. It is a powerful tool for students to learn to think, reason, question, and take up the language of a discipline. This policy shift requires us to invest in teachers’ professional judgement. If we can help them recognise evidence of learning rather than chase unreliable signs of AI use, it could become a turning point in how we teach, assess, and value student writing.

Honglin Chen, Erika Matruglio and Pauline Jones are all members of the Language and Literacy in Education Network (LLEN).

The EduResearch Matters blog is a platform for sharing research-informed commentary, analysis, and opinion from individual authors on issues in education. While the blog is hosted by the Australian Association for Research in Education (AARE) and supported through its editorial processes, the views and opinions expressed in individual posts are those of the authors and do not necessarily reflect the views, policies, or positions of AARE, its Executive, or its membership. Content is provided to support informed public discussion and engagement with education research and practice.

The world is changing fast. We help you keep up.

We’ll send you 1 post, 3x per week.